NY Tech Week 2026 — Agent Skill

You are working inside a community-maintained mirror of tech-week.com/calendar/nyc, restructured as flat markdown files so other agents can read, filter, and rank without scraping the live calendar.

Dataset shape

ny-tech-week-reference/
├── README.md             ← human-readable overview
├── SKILL.md              ← you are here
├── INDEX.md              ← chronological + by-host + by-neighborhood listing
├── HOST-STATS.md         ← top 50 hosts + recurring co-host pairs
├── manifest.json         ← raw calendar extraction
├── users.json            ← 2,406 resolved Partiful profiles (id → name, bio, photo, socials, tags)
├── events/               ← 1,410 markdown files, one per event
└── images/               ← 1,401 hero images (PNG/JPG), keyed by event_id

Each events/*.md has YAML frontmatter and a body. Filename is YYYY-MM-DD-HHMM-<slug>.md — sortable by start time.

Frontmatter quick reference

| Field | Type | Notes | |-------|------|-------| | title | string | From Partiful | | event_id | string | Partiful event ID (last segment of RSVP URL) | | date, day | string | ISO date + day name in ET | | start_time, end_time | string | "10:00am ET" — pretty form | | start_iso, end_iso | ISO 8601 | UTC, for date math | | host | string | Calendar's joined host string (up to 2 names visible) | | owner_count | int | True host count from Partiful (includes platform admin — see Quirks) | | owner_ids | liststring] | All Partiful user IDs; join with users.json for names + bios | | venue_name, venue_address | string | From Partiful's structured location | | neighborhood | string | NYC neighborhood like Chelsea, Midtown, Virtual (NYC) | | rsvp_url, public_short_url | URL | Partiful links | | google_maps | URL | Pre-formed maps query | | is_capped, max_capacity, remaining_capacity | bool / int | When the host capped the guestlist | | going_guest_count, total_guest_count, approved_guest_count | int | Guest count at crawl time (2026-05-30) | | at_capacity | bool | Whether RSVPs are currently closed | | canceled | bool | true if the host marked the event canceled on Partiful | | canceled_at | ISO 8601 | When cancellation was logged (only when canceled: true) | | canceled_by | string | Partiful user ID that triggered the cancellation | | cancellation_message | string | The host's note explaining why (verbatim, may be multi-line) | | guest_action | "APPLY" or "RSVP" | APPLY = hosts approve; RSVP = open | | visibility | string | "public" for the entire dataset | | badges | liststring] | ["Sponsored"], ["Morning"], etc. | | image | URL | Partiful Firebase image URL (may expire) | | local_image | string | Path to archived image in images/ (use this, not image) | | fetch_status | "ok" or "failed" | 7 events are stubs (Partiful 404'd at crawl time) |

Body shape

# {title}
![{title}](images/{event_id}.png)

**Hosts:** {calendar host string} _(+N more on Partiful)_
**When:** {pretty date}
**Where:** {venue · address · neighborhood}
**Capacity:** capped at N · X taken, Y remaining
**Going:** N of M
**Access:** apply / hosts approve
**RSVP:** {url}
**Map:** {google maps url}

### Hosts on Partiful

- [{name}](https://partiful.com/u/{id}) — _{bio}_
- [{name}](https://partiful.com/u/{id})
- _(plus Partiful platform admin auto-added to most events)_

## Description

{full Partiful description, verbatim}

---

[Apply on Partiful →]({url})

Quirks worth knowing

1. The platform admin user. 7DFu4rITofNzKIjA7hCx appears in owner_ids for ~99% of events (1,362 of 1,374). Treat it as noise — it's an automated TechWeek/Partiful account, not a real host. Filter at analysis time:

python real_hosts = [oid for oid in owner_ids if oid != "7DFu4rITofNzKIjA7hCx"] owner_count includes it to match Partiful's UI count.

1. Capacity fields only present when capped. If is_capped: false, max_capacity and remaining_capacity will be absent. Use going_guest_count / total_guest_count for uncapped events.

1. 7 stub events (fetch_status: failed) had their Partiful pages 404 at crawl time. Frontmatter is sparse, body is a stub with the RSVP URL. They still exist in events/ for completeness — handle by checking fetch_status first.

1. Calendar host string is truncated. host: "Foo, Bar" shows up to 2 hosts. For the true list, look at ### Hosts on Partiful in the body, or join owner_ids against users.json.

1. One outlier date. One event (Pitch and Rum) is dated 2026-06-11, outside the official June 1–7 window. It's on tech-week.com's calendar so it stays.

1. Cancellations stay in the dataset. When a host marks an event CANCELED on Partiful, we keep the file (so the URL stays resolvable, the data stays diff-able across crawls, and downstream agents can see what was cancelled and why) but flag it loudly: canceled: true in frontmatter, a ⚠ CANCELED banner at the top of the body with the host's cancellation_message. Filter at analysis time:

python live_events = [f for f in events if not f.get("canceled")]

1. Crawl provenance. Snapshotted 2026-05-30 (prior round 2026-05-21; see CHANGES-2026-05-30.md for the diff). Event details, RSVP availability, and guest counts may have drifted since.

Common queries

"What events are happening on Tuesday June 2 in SoHo?"

bash
grep -l 'date: 2026-06-02' events/*.md \
  | xargs grep -l 'neighborhood: "SoHo"'

"Which events focus on agent orchestration?"

bash
grep -l -i 'agent' events/*.md \
  | xargs grep -l -iE '(orchestrat|autonomous|multi-?agent|MCP)'

"Top 10 hosts by event count"

See HOST-STATS.md, or compute from events/*.md owner_ids joined with users.json:

python
import json, re, pathlib, collections
users = json.load(open("users.json"))
counts = collections.Counter()
for f in pathlib.Path("events").glob("*.md"):
    m = re.search(r'owner_ids:\s*(\[[^\]]+\])', f.read_text())
    if m:
        for uid in json.loads(m.group(1)):
            if uid != "7DFu4rITofNzKIjA7hCx":
                counts[uid] += 1
for uid, n in counts.most_common(10):
    print(n, users.get(uid, {}).get("name"))

"Build a shortlist for a specific person"

Score each event against the person's interests (use their bio, role, or stated themes as keyword seeds), then pick top N per day. See scripts/score.py for the pattern used to score events for Atin Woodard / Stage 11. Do not modify individual event files for a person-specific shortlist — write the picks to a new file like picks-for-<name>.md. Keep events/ neutral so it stays useful to everyone.

"Show me all cancelled events and why"

bash
grep -l '^canceled: true' events/*.md | while read f; do
  title=$(grep -m1 '^title:' "$f" | sed 's/title: //; s/^"//; s/"$//')
  msg=$(grep -m1 '^cancellation_message:' "$f" || true)
  printf "%s\n  %s\n  %s\n\n" "$title" "$f" "$msg"
done

"Find events at risk of filling up"

Where is_capped: true and remaining_capacity is small relative to max_capacity:

python
import re, json, pathlib
for f in pathlib.Path("events").glob("*.md"):
    s = f.read_text()
    if re.search(r"is_capped:\s*true", s):
        max_c = int(re.search(r"max_capacity:\s*(\d+)", s).group(1))
        rem = int(re.search(r"remaining_capacity:\s*(\d+)", s).group(1))
        if rem / max_c < 0.15:
            title = re.search(r'title:\s*"([^"]+)"', s).group(1)
            print(f"{rem}/{max_c} remaining: {title}")

Re-crawling

NYTW data shifts daily. To refresh:

1. Open the calendar in a browser; scroll through the full page (or use c11 / Playwright to script the scroll).
2. Run scripts/fetch.py against the regenerated manifest.
3. Run scripts/enrich.py to refresh capacity + image data.
4. Run scripts/resolve_hosts.py to re-resolve any new host IDs.
5. Run scripts/render_body.py to regenerate the rendered markdown.
6. Run scripts/build_index.py and scripts/host_stats.py.

Each step is idempotent — re-runnable without duplication.

What this dataset is not

• Not authoritative. Always include rsvp_url so humans can verify. tech-week.com is the source of truth; this repo is a convenience layer.
• Not real-time. Guest counts, capacity, and even event existence drift after the crawl date.
• Not opinionated. events/ is neutral. Person-specific picks belong in separate files.

Attribution

Event metadata © respective hosts and tech-week.com / a16z. Partiful images © Partiful + respective uploaders. This dataset is shared under fair-use for community planning and agent tooling.